DESCRIPTION: Dr. Abagyan has previously demonstrated that an improved energy function, combined with a powerful conformational search procedure, can accurately predict the detailed (all-atom) structure of small molecules and polypeptide fragments. It is proposed to apply these advances to the development of a comprehensive and accurate homology prediction procedure. Application of this method to homology modeling test cases showed that it permits successful prediction of loops and side chains under idealized conditions. However, the deformed yet rigid structural environment of modeled loops or other backbone fragments was found to be the largest obstacle to successful homology modeling. It is proposed: to adapt the loop modeling method to permit rearrangements in the loop's surroundings; to automate the determination of this "soft" environment; to apply the prediction method to polypeptide ends, mobile secondary structure elements and peptide plane orientation errors; to test and refine all components of the method, using an optimal test set of known homologous pairs from the standard database.